CrackML by @ml.with.umang
Interview questions / ML System Design
ML System Design interview question

Design Gun-Violence Video Moderation

Design an ML system to detect and moderate gun-violence content in short-form video.

hardsystem designEvidence 66/1001 source reportTikTok

The 60-second answer

Start from policy taxonomy and actions: depiction, threat, news/documentary context, severity, age/region rules, and false-positive vs false-negative costs. Combine sampled video frames, audio/ASR, OCR, metadata, and user/context signals in a staged multimodal pipeline with calibrated confidence.

Build the answer in this order

1
Frame the problem

Start from policy taxonomy and actions: depiction, threat, news/documentary context, severity, age/region rules, and false-positive vs false-negative costs.

2
Design the data path

Combine sampled video frames, audio/ASR, OCR, metadata, and user/context signals in a staged multimodal pipeline with calibrated confidence.

3
Choose the modeling stack

Route uncertain/high-impact cases to human review and preserve context so classifiers do not over-enforce benign reporting or educational content.

4
Serve, evaluate, iterate

Evaluate policy-weighted precision/recall by slice and monitor prevalence, appeals, reviewer disagreement, latency, adversarial edits, and policy drift.

A useful interview mental model

This is the shape of a strong answer—not a script to memorize.

01Requirements
02Data
03Model / Retrieval
04Serving
05Monitor

Senior-level signal

  • Senior answers separate policy correctness from visual detection accuracy and design for policy versioning.
  • Include adversarial robustness, multimodal disagreement, and safe rollback when enforcement rates shift.

What the interviewer is really testing

Product framing, data design, modeling choices, serving constraints, reliability, evaluation, and explicit trade-offs.

Likely follow-up questions

What changes at 10× traffic or data volume?
Which failure mode would you monitor first in production?
How would you evaluate this offline and online before rollout?

Common weak-answer patterns

  • Jumping to a model before defining the product contract.
  • Listing components without bottlenecks, metrics, or failure handling.
  • Ignoring data quality, serving latency, monitoring, and iteration.